Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 9 Distribution Pr 0 α = 1 α = 2 α = 3 α = 4 α = 5 DET 0 0 0 0 0 UNI 0.5 0 0 0 0 EXP 0.3679 0.1353 0.0498 0.0183 0.0067 PAR (k=1/3) 0.2963 0.0787 0.0315 0.0156 0.0089 Table 3. Empty probability vs. scaling factor under different traffic distributions. number of simultaneous wake-ups. Since γ j s are integers, we can compute the least common multiple (LCM) for all the elements of each Γ ∗ candidate. Note that the LCM gives the minimal number of BIs for which two or more clients wake up simultaneously. Therefore, a larger LCM means a smaller number of simultaneous wake-ups. We therefore choose the best Γ based on the largest LCM and denote it as Γ ∗ i . In the second sub-step, given (β i , Γ ∗ i ), i = 1, ,n + 1, we select the best Γ from the Γ ∗ i s that minimizes simultaneous wake-up. The criterion is based on the largest spread of their elements from one another which is measured by the ratio of the standard deviation and the mean of the elements in Γ ∗ i . Therefore, Γ ∗ is given by the Γ ∗ i that gives the highest ratio, and β ∗ is the corresponding β i . Step 3 (Determining Θ ∗ ) The final step determines Θ ∗ based on the Γ ∗ obtained in the last step. The motivation is to mitigate the possible unfairness in the frame buffering delay experienced by the clients. We assign a smaller θ ∗ j to the client with a larger γ ∗ j . In this way, the client that wakes up less frequently will have a higher priority to retrieve its frames during channel contention. To do so, we assign the default value to s j if its γ ∗ j is the largest (i.e. θ ∗ j = 31). We then increase other clients’ θ j by  Θ when their γ ∗ j decrease. They are given by θ ∗ j = 31 +  Θ (max ∀j (γ ∗ j ) − γ ∗ j ). 5.2 The optimal wake-up schedule Besides the main algorithm, the AP in C-PSM may also obtain the optimal wake-up schedule (WS). This optional step is to schedule the first wake-up times of the clients, so that the maximal number of waking clients at one BI epoch is minimized. The optimization problem is given by min r : max ν=0,1,2, N(ν, r, Γ ∗ ). (3) The vector r presents the sequence of the first wake-up times where the client s j first wakes up at the r j th BI epoch and r j ∈ [0, LCM(Γ ∗ ) − 1] is an integer. The function N(ν, r, Γ ∗ ) is the number of waking clients at the νth BI epoch when the clients wake up according to r and LI ∗ . WS consists of the optimal solution denoted as r ∗ which will further decrease the simultaneous wake-ups if two or more elements of Γ ∗ are the same or have the same common factor. Since the optimization problem (3) can be decomposed into a series of wake-up scheduling problems (WSPs) (Lin et al., 2006), we solve it by developing an algorithm based on the stepwise solving method for WSP. When a new PSM-enabled client s j joins an infrastructure network (including a set of m clients, S)intheνth BI epoch, WSP is formulated to minimize the maximal number of wake-up clients 11 Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 10 Will-be-set-by-IN-TECH in the following BI epoches. The optimization problem of WSP is given by min k j (ν) : max u=1,2, {N(ν + u)} (4) where N (ν + u) is the number of waking clients at the (ν + u)th BI epoch and the wake-up counter k j (ν) records the remaining BIs that the client s j will wake up. Moreover, N(ν + u) equals to ∑ i∈S∪j w i (ν + u) where the wake-up indicator w i (ν + u) is 1 if s i wakes up at the (ν + u)th BI epoch; otherwise, it is 0. Given the LI parameter of each client γ i ,0≤ k i (ν) ≤ γ i − 1, i ∈S∪j, the stepwise solving method (Lin et al., 2006) can calculate k ∗ j (ν), the optimal wake-up counter of s j . And k ∗ j (ν) is a function f (j, γ j , m, w i (ν), k i (ν), γ i ), i = 1, . . . , m.Itis easy to see that k ∗ j (ν) is the optimal first wake-up time of s j when ν = 0, i.e. r ∗ j = k ∗ j (0). According to Γ ∗ , our algorithm obtains k ∗ j (0) for each client at ν = 0. Therefore, we determines WS as r ∗ =[k ∗ 1 (0), ,k ∗ c (0)]. The client s j first wakes up optimally at the r ∗ j th BI epoch, ∀j = 1, . . . , c. For example, if r ∗ j = 1, s j will miss the first beacon frame at the beginning of simulation but wake up for the first time after β ∗ . The detail steps of our algorithm for obtaining r ∗ is given below: 1. Initialize the following variables: ν = 0, S = {s 1 }, m = 1, w 1 (ν)=1, k 1 (ν)=0, r ∗ 1 = 0 and j = 2. 2. If j > c, return r ∗ and exit; else, go to step 3. 3. Find the optimal wake up time of s j , where k ∗ j (ν)=f (j, γ ∗ j , m, w i (ν), k i (ν), γ ∗ i ), i = 1, . . . , m. 4. Update variables: r ∗ j = k ∗ j (ν), m = m + 1 and S = S∪s j . 5. If r ∗ j = 0, then w j (ν)=1; else, w j (ν)=0. 6. Increase j by 1 and go back to step 2. 6. Performance evaluation We evaluate the performance of C-PSM and compare it with the PSM with default parameters (which is referred to as standard PSM or S-PSM). We do not compare C-PSM with other user-centric/AP-centric PSM schemes, because the design objectives and the study scopes are different. For example, C-PSM improves energy efficiency for all clients, whereas the user-centric PSM schemes consider only a single client. On the other hand, C-PSM is standard-compliant, but most AP-centric schemes are not compatible with the PSM scheme. 6.1 Evaluation methodology In order to evaluate the effectiveness of different components of C-PSM, we examine three different versions. The first one is a “full version” which includes the optional optimal wake-up sequence discussed in the last section. The other two, Scheme-1 and Scheme-2, on the other hand, exclude this option and adopt the default congestion window size. Moreover, Scheme-2 adopts the default Γ value. To sum up, we compare the following schemes in reference to S-PSM. 1. C-PSM: β ∗ , Γ ∗ , Θ ∗ , r ∗ ; 12 Energy Technology and Management Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 11 2. Scheme-1: β ∗ , Γ ∗ , θ j = 31, r j = 0; 3. Scheme-2: β ∗ , γ j = 1, θ j = 31, r j = 0; 4. S-PSM: β = 100ms, γ j = 1, θ j = 31, r j = 0. We use the following four performance indices for comparing C-PSM, Scheme-1, and Scheme-2 against S-PSM: power saving (index η P ), throughput (index η T ), energy efficiency (index η T/P ), and frame buffering delay (index η D ). The notations with a superscript S − PSM refers to S-PSM, whereas that without refer to C-PSM, Scheme-1, or Scheme-2. For easy comparison, a positive value indicates improvement over S-PSM. η P =(P S−PS M − P)/P S−PS M × 100%, η T =(T − T S−PS M )/T S−PS M × 100%, η T/P =(R T/P − R S−PS M T/P )/R S−PS M T/P × 100%, η D = 1 c × c ∑ j=1 (d S−PS M j − d j )/d S−PS M j × 100%. 6.2 Two clients We first evaluate C-PSM in the two-client system. Given Δ =[15; 25]ms, the AP obtains the optimal parameters of C-PSM under different traffic distributions, as shown in Table 4. Δ(ms) distribution β ∗ (ms) Γ ∗ Θ ∗ r ∗ [15;25] DET 10 [2;3] [39;31] [0;0] UNI 26 [1;2] [39;31] [0;0] EXP,PAR 38 [1;2] [39;31] [0;0] [20;30;30] DET 16 [1;2;2] [39;31;31] [0;0;1] UNI 30 [1;2;2] [39;31;31] [0;0;1] EXP,PAR 46 [1;2;2] [39;31;31] [0;0;1] Table 4. Optimal parameters of C-PSM ( β = 2ms and  Θ = 8). (a) Total power, P. (b) Total energy efficiency, R T/P . Fig. 3. A comparison of the four PSM schemes with Δ =[15; 25]ms. 13 Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 12 Will-be-set-by-IN-TECH Figure 3 depicts that C-PSM outperforms S-PSM on saving energy and improving energy efficiency under different distributions. C-PSM achieves lowest P and highest R T/P among the four schemes. Scheme-1 performs a little worse than C-PSM, since it does not adopt Θ ∗ . Comparing with scheme-1, scheme-2 increases power and decreases energy efficiency, since it does not use Γ ∗ . Adopting β ∗ , Scheme-2 still outperforms S-PSM under all traffic distributions except the DET distribution. Scheme-2 is the worst under deterministic traffic, because two clients wake up every β ∗ = 10ms and they both waste energy on the frequent unnecessary wake-ups. Note that, the WS is not adopted, since r ∗ is a zero vector when γ j (∀j) are relative prime. In this case, all clients wake up at the beginning of simulation. index, % scheme DET UNI EXP PAR η P C-PSM 25.41 28.75 29.73 28.13 Scheme-1 24.91 27.28 27.53 26.47 Scheme-2 -20.82 17.52 21.10 16.48 η T/P C-PSM 34.63 41.18 43.01 39.76 Scheme-1 33.71 38.33 38.65 36.57 Scheme-2 -16.88 21.95 27.38 20.30 η D C-PSM 82.33 68.79 54.80 54.18 Scheme-1 82.08 68.15 53.07 53.56 Scheme-2 94.54 79.79 69.88 68.75 η T C-PSM 0.41 0.59 0.50 0.45 Scheme-1 0.40 0.60 0.48 0.42 Scheme-2 0.43 0.58 0.50 0.47 Table 5. Indices of C-PSM, Scheme-1 and Scheme-2, Δ =[15; 25]ms. As shown in Table 5, all indices of C-PSM are positive and the improvements of C-PSM over S-PSM are significant under different distributions. For example, compared with S-PSM, C-PSM reduces power consumption by 29.37%, improves energy efficiency by 43.01% and reduces average buffering delay by 54.8% under the EXP distribution of traffic. We also find that C-PSM has largest η P and η T/P . That is, C-PSM which employs β ∗ , Γ ∗ and Θ ∗ together, performs the best in saving power and increasing energy efficiency. In C-PSM, the benefit of adopting β ∗ and Γ ∗ is significant whereas the improvement due to Θ ∗ is minor. Scheme-1 using β ∗ and Γ ∗ has obtained large positive indices. Its indices are slightly less than C-PSM’s indices. For example, the energy efficiency is improved by 38% while the η T/P of C-PSM is 43.01%. Therefore, the usage of Θ ∗ is helpful to save energy but not much. Moreover, β ∗ and Γ ∗ jointly play the major roles in improving PSM performance. In contrast, the PSM performance greatly degrades without using Γ ∗ . The η P and η T/P of Scheme-2 are much smaller than the ones of C-PSM and Scheme-1. Scheme-2 therefore is worse than these two schemes. It is even worse than S-PSM, because of its negative indices under the DET distribution of traffic. Next, we compare C-PSM and S-PSM under the EXP distribution of traffic. As shown in Figure 4(a), the AP buffering delay is shortest in C-PSM, whereas it is longest in S-PSM. Consequently, the clients using C-PSM can return to sleep mode earlier, because the frames with shorter buffering delay are retrieved faster than those in S-PSM. Moreover, C-PSM improves the fairness of clients, since it greatly decreases the delay difference of the two clients. C-PSM speeds up the retrieval of frames in the fast client, because the delay of s 1 is one sixth of that in S-PSM. At the same time, it does not degrade the slow client, since the 14 Energy Technology and Management Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 13 (a) The AP buffering delay. (b) The ratios of unnecessary wake-ups and simultaneous wake-ups. (c) The collision ratios. Fig. 4. A comparison of the four schemes under the EXP distribution with Δ =[15; 25]ms. delay of s 2 reduces a little. Figure 4(b) shows that C-PSM greatly reduces the chances that two clients simultaneously wake up to compete for channel, since its R bB/B,2 is lowest. In contrast, S-PSM lets two clients wake up simultaneously at a large proportion of BI epochs, since its R bB/B,2 is near 1. The clients using C-PSM spend less time on channel contention, consume less energy on idle mode and then achieve higher energy efficiency. Figure 4 illustrates that C-PSM and Scheme-1 have similar performance metrics. The collision ratios of PS-Poll and data frames in C-PSM are less than those in Scheme-1. It means that Θ ∗ is useful for reducing channel collisions. That is why C-PSM performs better than Scheme-1 with slightly higher indices. Scheme-2 outperforms S-PSM under the EXP distribution of traffic, since it achieves shorter AP buffering delay and less simultaneous wake-up ratio than S-PSM. Scheme-2 is less energy-efficient than C-PSM and Scheme-1, because it spends more energy on unnecessary wake-ups and channel contention. The clients in Scheme-1 frequently wake up every β ∗ but nearly 20% of wake-ups are unnecessary while the unnecessary wake-ups ratio is less than 10% in C-PSM. Without using Γ ∗ , the clients in Scheme-2 spend more energy on idle mode 15 Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 14 Will-be-set-by-IN-TECH when they simultaneously wake up to compete channel with a higher probability. Shown in Figure 4(b), they are involved in channel contention at 68% of BIs, i.e. R bB/B,2 = 68% while the R bB/B,2 in C-PSM is only 40%. Furthermore, the collision ratios in Scheme-2 are higher than those in C-PSM and Scheme-1, shown in Figure 4(c). For example, the PS-Poll collision ratio in Scheme-2 is the highest, nearly 1.5 times of that in C-PSM. Therefore, the clients in Scheme-2 have to spend more energy to handle the collisions. On the other hand, Scheme-2 has shorter delay for the slow client s 2 than C-PSM and Scheme-2. However, its benefit is too small to affect the performance. Additionally, the collision ratios of ACK are almost zero in Figure 4(c). The reason is that an awaken client always returns ACK after it has finished receiving a data frame and a SIFS has elapsed. The channel is rarely occupied by the other client or the AP within such a short SIFS. Therefore, ACKs rarely suffer from collisions especially when c = 2. If the number of clients increases, the probability of ACK collisions will increase as the channel contention intensifies. 6.3 More than two clients We have applied C-PSM to a network with more than two clients. Firstly, we evaluate C-PSM when the number of clients is 3 and two clients have the same mean of inter-frame arrival times. According to Δ =[20; 30; 30] where ρ = 13.18% < 30%, the main algorithm obtains the optimal parameters of C-PSM, also shown in Table 4. Note that when the elements of Γ ∗ are not relative prime numbers or even the same, C-PSM must adopt WS. In this case, γ ∗ 2 and γ ∗ 3 have a greatest common divisor 2, and then C-PSM uses WS, i.e., r ∗ =[0; 0; 1]. s 1 and s 2 wake up at the beginning of simulation while s 3 defers the first wake-up time for one BI. index, % scheme DET UNI EXP PAR η P C-PSM 36.38 39.08 36.78 36.31 C-PSM not WS 29.00 30.08 26.43 27.33 η T/P C-PSM 59.86 65.92 59.11 58.00 C-PSM not WS 43.22 44.56 36.73 38.41 η D C-PSM 84.00 68.69 52.16 51.98 C-PSM not WS 81.80 64.62 45.23 46.37 η T C-PSM 1.71 1.08 0.60 0.63 C-PSM not WS 1.69 1.07 0.59 0.58 Table 6. Indices of C-PSM with/without WS, Δ =[20; 30; 30]ms. The positive indices in Table 6 show that C-PSM outperforms S-PSM in terms of power saving, energy efficiency and AP buffering delay while keeping or slightly increasing throughput in the three-client network. For example, C-PSM reduces power consumption by 36.78%, improves the energy efficiency by 59.11% and shortens the average buffering delay by 52.16% under the EXP distribution of traffic while total throughput remains almost the same. On the other hand, the indices of C-PSM without WS are less than the ones of C-PSM under all traffic distributions. For example, without using WS, the η T/P Of C-PSM is decreased at most by 22% under the EXP distribution of traffic. Therefore, WS is much helpful to improve energy efficiency when the symmetric clients exist. Next, we compare the simulation results of C-PSM, C-PSM without WS and S-PSM to explain the above findings. As an example, we study these three schemes under the EXP distribution of traffic. 16 Energy Technology and Management Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 15 metrics C-PSM C-PSM not WS S-PSM P (Watt) 0.8242 0.9591 1.3037 T (10 5 bps) 4.7538 4.7536 4.7257 R T/P (10 5 bpJ) 5.7675 4.9563 3.6248 d 1 (ms) 34.4 36.3 234.2 d 2 (ms) 55.8 63.2 84.5 d 3 (ms) 54.9 64.7 87.4 R c/t 1.44% 1.78% 2.14% R u/w 10.06% 10.66% 4.86% R bB/B,2 83.83% 10.47% 7.64% R bB/B,3 0 39.05% 92.29% Table 7. A comparison of three schemes under the EXP distribution with Δ =[20; 30; 30]ms. C-PSM saves energy by shortening the period of channel contention, shown in Table 7. All the clients’ frame buffering delays of C-PSM are smaller than those of other two schemes. That is, each client can receive its buffered frames most quickly and then enter to sleep instead of spending much energy and time on idle mode during channel contention. C-PSM also saves energy by reducing channel contentions. It totally avoids all-client simultaneous wake-ups and R bB/B,3 is zero. On the other hand, in S-PSM, three clients wake up together to receive data in almost all BIs and R bB/B,3 is as high as 92.29%. At the same time, C-PSM consumes a small amount of energy on unnecessary wake-ups, since the total ratio of unnecessary wake-ups R u/w is near 10%. It also decreases the channel collisions where the total ratio of collisions R c/t is reduced by about one-third. From what has been discussed above, C-PSM outperforms S-PSM. C-PSM without WS obviously outperforms S-PSM but is worse than C-PSM. Without using WS, the total power increases, the total energy efficiency decreases and three awaken clients compete for receiving data in 39.05% of BIs. However, C-PSM can totally avoid the situation of all-client simultaneous wake-ups. On the other hand, the R bB/B,2 in the C-PSM without WS is smaller than the one in C-PSM. It is helpful to save energy but does not determine the energy consumption of all clients. The reason is that more energy is consumed on channel contention when all of three clients wake up simultaneously. In a global view, C-PSM saves more energy and achieves higher energy efficiency after using r ∗ . Therefore, WS is helpful to improve energy efficiency, because the number of clients which wake up at the same beacon epoch has been minimized. Furthermore, we find that C-PSM is applicable for a large scale network and saves more energy when the number of clients increases. These two sets of simulations evaluate the performance of our scheme when the number of PSM-enabled clients increases up to 20. In the first set of simulations, we let δ j = 10c(ms), j = 1, ,c. The total amount of traffic of all symmetric clients will not change with c, and the total arrival rate of frames λ always equals 100 frames per second. The traffic workload is light, when ρ = 11.3% < 30%. Under light traffic, S-PSM can save energy, since the average power of each client is always lower than its idle power. For example in Figure 5, P S−PS M is much less than the c times of client’s idle power under the EXP distribution of traffic. C-PSM scheme can further reduce energy consumption, since the power difference between P S−PS M and P C−PS M is always larger than zero. Moreover, this power difference increases with c. Therefore, C-PSM saves more energy when the number of clients increases. 17 Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 16 Will-be-set-by-IN-TECH Fig. 5. Total power verses c under the EXP distribution with δ j = 10c(ms). Index,% T j c=2 c=4 c=8 c=12 c=16 c=20 η P DET 22.30 63.75 79.20 71.54 75.52 72.98 UNI 45.37 72.52 78.82 70.58 71.74 72.22 EXP 51.09 70.33 76.07 70.98 72.27 72.14 PAR 50.76 70.43 76.68 69.77 70.91 70.33 η R T/P DET 29.07 177.76 396.05 263.65 325.31 286.14 UNI 83.50 265.28 384.96 251.95 270.96 281.65 EXP 105.04 238.69 327.07 257.23 277.64 281.02 PAR 103.89 239.39 338.50 241.89 260.80 255.76 η D DET 92.35 87.63 85.78 87.47 90.18 89.21 UNI 73.94 70.37 84.07 85.57 87.28 88.11 EXP 64.09 70.89 82.68 85.02 87.04 88.43 PAR 62.93 69.48 82.25 85.78 87.10 88.19 η T DET 0.29 0.70 3.17 3.51 4.10 4.32 UNI 0.24 0.40 2.72 3.55 4.83 6.02 EXP 0.29 0.48 2.19 3.68 4.73 6.14 PAR 0.40 0.37 2.27 3.36 4.94 5.55 Table 8. The indices of C-PSM verses the number of clients when δ j = 10c(ms), j = 1, ,c. Table 8 has shown that C-PSM achieves significant improvements of four main metrics in a large network. For example, the indices of η P , η T/P and η D in are as high as 76.07%, 327.07% and 82.68% individually when c = 8 under the EXP distribution of traffic. C-PSM scheme also improves clients’ throughput in a large network, since η T increases slightly with the number of clients. In the second set of simulations, we let δ j = 10 + 5j(ms), j = 1, ,c. The total packet arrival rate λ increases with the number of clients where ρ increases from 13.18% to 49.51%. When c ≥ 8, the traffic is not light any more, since ρ ≥ 32%. We firstly find that C-PSM has a wider applicability than the standard PSM. It is effective to save energy when the network supports many clients whose workload is not light. For 18 Energy Technology and Management Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 17 Fig. 6. Total power verses c under the EXP distribution with δ j = 10 + 5j(ms), j = 1, . . . , c. example in Figure 6 under the EXP distribution of traffic, P S−PS M is much close to the total power of c idle clients, when c ≥ 8. It is obvious that these clients are too busy to sleep and S-PSM cannot save much energy. After using C-PSM, P C−PS M is much less than the total power of c idle clients even when c increases to 20. That is, C-PSM is still effective to save energy even when the network workload is as high as ρ ≈ 50%. Moreover, C-PSM saves more energy when the number of clients increases, since that the power difference P S−PS M − P C−PS M increases with c. Although not shown here, the similar simulation results are obtained under other traffic distributions. Index,% T j c=2 c=4 c=8 c=12 c=16 c=20 η P DET 1.01 29.04 42.17 44.31 51.49 52.71 UNI 21.93 37.81 44.29 37.34 51.98 50.97 EXP 28.98 39.73 44.20 36.01 49.44 43.89 PAR 25.64 37.59 44.19 35.65 48.85 43.49 η R T/P DET 0.55 55.54 132.72 174.52 248.45 286.48 UNI 28.79 76.14 138.29 137.83 245.38 255.72 EXP 41.51 75.73 124.41 115.23 201.97 172.88 PAR 35.07 71.90 126.80 114.65 203.60 173.49 η D DET 96.60 94.98 95.59 93.39 94.00 91.36 UNI 76.05 79.90 85.27 85.56 86.88 85.72 EXP 66.12 69.74 77.41 77.21 79.52 76.21 PAR 65.05 69.82 77.69 77.10 79.66 76.30 η T DET 0.45 10.38 34.58 52.88 69.04 82.76 UNI 0.55 9.54 32.75 49.03 65.84 74.41 EXP 0.49 5.91 25.23 37.74 52.68 53.12 PAR 0.43 7.29 26.58 38.13 55.28 54.56 Table 9. The indices of C-PSM verses the number of clients when δ j = 10 + 5j(ms), j = 1, . . . , c. 19 Centralizing the Power Saving Mode for 802.11 Infrastructure Networks 18 Will-be-set-by-IN-TECH Table 9 also shows that C-PSM scheme outperforms S-PSM on saving power, improving energy efficiency, shortening delay and increasing throughput. When the traffic is not light (i.e. c ≥ 8), C-PSM improves energy efficiency a lot, since it not only reduces power consumption but also increases throughput greatly. For example, compared with S-PSM, C-PSM saves 49.44% of power, increases 52.68% of throughput and then finally achieves 201.97% higher energy efficiency in the sixteen-client system under the EXP distribution of traffic. 6.4 Effects of power consumption model on C-PSM The power profile of wireless device has a great impact on the performance of energy-saving scheme using sleeping (Nedevschi et al., 2008). This profile includes the power consumption of client in transmission, reception, idle mode, sleeping mode and mode transition (when the client wakes up from sleeping mode to active mode), as well as the wake-up time. Additionally, the energy consumed on client’s wake-up is the product of wake-up power 3 and wake-up time. The set of these above parameters are defined as a power consumption model in this chapter. We adopt model A (Feeney & Nilsson, 2001; Margi, 2006) in our simulator, which is widely used. Model A is comparable to the hardware characteristics of many popular wireless interface cards. The ratio of transmission power to reception power in model A is approximately 160% which is similar to the ratioes of ORiNOCO 11a/b/g ComboCard (Proxim Wireless Corporation, 2006a), ORiNOCO 11a/b/g PCI card (Proxim Wireless Corporation, 2006b), CISCO AIRONET 802.11A/B/G Wireless Cardbus adapter (Cisco Systems, Inc., 2004), CISCO AIRONET 350 Series Wireless LAN Client Adapters (Cisco Systems, Inc., 2005) and Aironet’s PC4800 PCMCIA NIC (Ebert et al., 2002). The reception power is near to the idle power in model A which is the same in CISCO AIRONET 802.11A/B/G Wireless Cardbus adapter (Cisco Systems, Inc., 2004) and Aironet’s PC4800 PCMCIA NIC (Ebert et al., 2002). Moreover, the sleep power is about an order of magnitude lower than the idle power in model A. The ratio of idle power to sleep power denoted as R I/S equals to 1167% which is common in many popular wireless network interface cards. State model A model B model C model D model E Transmission power 1.4W 1.65W 0.75W 1.3W 0.85W Reception power 0.9W 1.4W 0.75W 0.95W 0.85W Idle power 0.7W 1.15W 0.75W 0.79W 0.85W Sleeping power 0.06W 0.045W 0.05W 0.17W 0.005W R I/S 1167% 2556% 1500% 468% 17000% Wake-up power 0.7*2W 1.15*2W 0.75W 0.51W 0.85*2W Wake-up Time 2ms 2ms 2ms 13ms 2ms Wake-up Energy 0.003J 0.005J 0.0015J 0.0066J 0.0034J Table 10. Five power consumption models. Next, we compare model A with other power consumption models model B (Jung & Vaidya, 2002; Simunic et al., 2000), model C (Anastasi et al., 2007; Krashinsky & Balakrishnan, 2005), model D (Jeong et al., 2004) and model E(C-Guys, Inc., 2004) listed in Table 10. In order to study 3 During the mode transition, the client’s power consumption is near or higher than transmission power (Stemm & Katz, 1997). It could be estimated as two times of idle power (Jung & Vaidya, 2002), for example the models A, B and E. 20 Energy Technology and Management [...]... PAR ηRT/P DET UNI EXP PAR A 26 .16 27 . 72 28.99 29 .56 36 .25 39.01 41 .29 42. 30 B 30.86 32. 32 33. 72 34.37 45. 52 48.46 51.37 52. 71 C 46.46 42. 12 39.63 40.31 87.91 73.58 66.17 67.93 D -4.86 5.65 12. 69 13. 02 -4.05 6.48 14.91 15 .24 E 38.44 38.73 39.58 40.30 63.44 63.98 66.04 67.91 Table 11 Indices of C-PSM in different power consumption models when c = 2, Δ = [15; 25 ] 7 Conclusions and future works We propose... force F and ymax and the relationship between displacement s and maximum distance ymax, the bending energy can be express by formula 3 For the given material (with a modulus E and a height l) bent to a displacement compression (s), the bending energy Ubend is affected directly by the inertia moment (I) In a word, Ubend is in a function of I F2 l (l − x )2 dx 2EI 0 F2 l ( x − l )2 dx = 2EI 0 F2 1 =... x )2 dx 2EI 0 F2 l ( x − l )2 dx = 2EI 0 F2 1 = ⋅ ( x − l )3 0 l 2EI 3 F 2l3 = 6EI  Ubend = and F=− 3EIy max l3 30 Energy Technology and Management ∴ U bend = 3EI ( ymax )2 2l3 = 5E ⋅ I ⋅ s 2l2 (3) An example about the effect of cross section geometry on I is given in Fig.4 For a given material with the same section area (A:50mm2) and thickness (t:4mm), making the structure to have cross section... 8 02. 11 Infrastructure for 8 02. 11 Infrastructure Networks 23 21 Lee, J., Rosenberg, C & Chong, E (20 06) Energy efficient schedulers in wireless networks: design and optimization, Mobile Networks and Applications 11(3) Lei, H & Nilsson, A (20 07) Queuing analysis of power management in the IEEE 8 02. 11 based wireless lans, IEEE Transactions on Wireless Communications 6(4) Lin, H., Huang, S & Jan, R (20 06)... Jon, C & Andrew, R (20 10) Exhausting battery statisticsunderstanding the energy demands on mobile handsets, Proc MobiHeld Nath, S., Anderson, Z & Seshan, S (20 04) Choosing beacon periods to improve response times for wireless HTTP clients, Proc the ACM International Workshop on Mobility Management and Wireless Access (MobiWac) Nedevschi, S., Popa, L & Iannaccone, G (20 08) Reducing networking energy consumption... http://www.j-sim.org/ 24 22 Energy Technology and Management Will-be-set-by-IN-TECH Wang, J F., Fang, Y G & Wu, D P (20 06) A power-saving multi-radio multi-channel mac protocol for wireless local area networks, Proc IEEE INFOCOM Zeng, Z., Gao, Y & Kumar, P R (20 11) Sofa: A sleep-optimal fair-attention scheduler for the power-saving mode of wlans, Proc IEEE ICDCS Zhang, F & Chanson, S (20 03) Throughput and value... Fiber-Reinforced Plastic (FRP) Tubes as Energy Absorption Element in Vehicles 29 delaminations; Ubend is the energy of the bending of fronds; Uff is the energy required for fiber fracture and Ufr is the energy associated with friction Because these fractures occur simultaneously and they correlate and affect each other, it is difficult to evaluate individual energy absorption contribution and construct a design criterion... infrastructure-mode 8 02. 11 wireless LANs, Computer Communications 29 : 3483–34 92 Margi, C (20 06) Energy Consumption Trade-offs in Power Constrained Networks, PhD thesis, University of California Santa Cruz MATLAB Central (20 03) IEEE 8 02. 11a WLAN model, http://www.mathworks.com/ matlabcentral/fileexchange/3540 MATLAB Central (20 09) 8 02. 11b PHY matlab code, http://www.mathworks.com/ matlabcentral/fileexchange/ 321 3/ Narseo,... in crashes In particular, a tubular component, termed “Front Side Member”(Fig 1), is designed to be installed behind the 28 Energy Technology and Management bumper as a special kind of energy absorption element which can absorb most of the impact energy through the fracture of itself in the chapter in particularly For a FRP tube which was fractured through progressive crushing mode, the energy absorbing... B & Wolisz, A (20 02) Measurement and simulation of the energy consumption of an WLAN interface, Technical report, Technical University Berlin, Telecommunication Networks Group Feeney, L & Nilsson, M (20 01) Investigating the energy consumption of a wireless network interface in an ad hoc networking environment, Proc IEEE INFOCOM Gao, Y., Zeng, Z & Kumar, P R (20 10) Joint random access and power selection . PAR η P C-PSM 25 .41 28 .75 29 .73 28 .13 Scheme-1 24 .91 27 .28 27 .53 26 .47 Scheme -2 -20 . 82 17. 52 21.10 16.48 η T/P C-PSM 34.63 41.18 43.01 39.76 Scheme-1 33.71 38.33 38.65 36.57 Scheme -2 -16.88 21 .95 27 .38 20 .30 η D C-PSM. 28 6.14 UNI 83.50 26 5 .28 384.96 25 1.95 27 0.96 28 1.65 EXP 105.04 23 8.69 327 .07 25 7 .23 27 7.64 28 1. 02 PAR 103.89 23 9.39 338.50 24 1.89 26 0.80 25 5.76 η D DET 92. 35 87.63 85.78 87.47 90.18 89 .21 UNI 73.94. 79 .20 71.54 75. 52 72. 98 UNI 45.37 72. 52 78. 82 70.58 71.74 72. 22 EXP 51.09 70.33 76.07 70.98 72. 27 72. 14 PAR 50.76 70.43 76.68 69.77 70.91 70.33 η R T/P DET 29 .07 177.76 396.05 26 3.65 325 .31 28 6.14 UNI